Data-Based Estimation of Operating Behavior of an Mr Device

1. A method for computer-aided estimation of an operating behavior for an MR device having a set of device components, comprising: providing a memory with a set of digital models, wherein each digital model simulates an operating behavior of a respective component of the MR device, and the digital models are interconnected in accordance with the structure and/or functionality of the MR device to form a higher-order model which simulates the operating behavior of the MR device; acquiring operating data of the MR device; and during an inference phase: (i) accessing, by a processor, the memory storing the acquired operating data, and for each digital model, generating a component-specific intermediate result that represents a simulated operating behavior of the corresponding device component; (ii) comparing, by a comparator module, the component-specific intermediate results of the individual digital models against one another across the device components; and (iii) determining, based on the comparison, a result to estimate the operating behavior of the MR device; and outputting a result.

2. The method as claimed in claim 1, wherein the comparator module is configured to use timestamp data with the operating data for comparison.

3. The method as claimed in claim 1, further comprising: generating synthetic operating data using a Markov model in a training phase, wherein the synthetic operating data is generated based on historical operating data of the MR device; and training the digital models and/or the higher-order model using the synthetic operating data to predict future operating behavior of the MR device.

4. The method as claimed in claim 1, wherein the digital models and/or the higher-order model comprises a trained long short-term memory (LSTM) neural network.

5. The method as claimed in claim 1, further comprising: training each of the digital models individually from the set of digital models.

6. The method as claimed in claim 1, wherein the cross-component model is a higher-order cross-component model that is self-learning in that the estimated operating behavior is compared with a real operating behavior resulting from measured operating data to train the higher-order cross-component model in the event of deviation.

7. The method as claimed in claim 1, wherein the operating data comprises at least timestamp data and/or error messages, and the timestamp data is transformable according to a transformation rule.

8. The method as claimed in claim 1, further comprising: performing a pattern recognition algorithm on the acquired operating data and/or the estimated operating data, for gradient coil related operating parameters, for scan sequence related parameters, for excitation pulses, and/or for an operating temperature of coils.

9. The method as claimed in claim 1, wherein the acquisition of operating data is triggered automatically during operation of the MR device and is performed by readout from an event log file.

10. A non-transitory computer program product comprising a computer program code for carrying out a method as claimed in claim 1 when the computer program is executed on an electronic device, a server, and/or computer.

11. A server system for operating an MR device farm comprising a set of distributed MR devices, the server system comprising: data interfaces to the MR devices; a model interface to the memory; and a processor configured to carry out the method as claimed in claim 1.